create-meta-prompts

About This Skill

This skill extends shared/template-generator for common creation workflows.

Meta-prompt-specific focus: Purpose-driven patterns, folder structure, output organization, and chaining.

What Are Meta-Prompts?

Meta-prompts are designed for pipelines where one Claude output becomes input for another:

• Research → Gather information or understand something
• Plan → Create an approach, roadmap, or strategy
• Do → Execute a task and produce artifacts
• Refine → Improve existing research or plan outputs

Each stage produces outputs optimized for the next stage plus human-readable summaries.

Purposes and Output

Research

Goal: Understand a topic deeply

Output structure:

• XML-formatted analysis with findings
• Key insights and patterns
• Data supporting conclusions
• References and sources

Consumed by: Plan or Do prompts

Plan

Goal: Create strategy or roadmap

Output structure:

• Phase-by-phase approach
• Dependencies and ordering
• Success criteria for each phase
• Alternative approaches

Consumed by: Do prompts for implementation

Do

Goal: Execute and create artifacts

Output structure:

• Created files/code
• Execution decisions and rationale
• SUMMARY.md with completion status
• Files produced with locations

Consumed by: Human or next Do prompt

Refine

Goal: Improve existing research/plans

Output structure:

• Updated output with improvements
• Rationale for changes
• Version history in archive/
• SUMMARY.md comparing versions

Consumed by: Do prompts or further refinement

Folder Organization

Meta-prompts live in .prompts/ with numbered, purpose-named folders:

.prompts/
├── 001-auth-research/
│   ├── completed/
│   │   └── 001-auth-research.md
│   ├── auth-research.md       # Output (XML for Claude)
│   └── SUMMARY.md             # Executive summary (human)
├── 002-auth-plan/
│   ├── completed/
│   │   └── 002-auth-plan.md
│   ├── auth-plan.md
│   └── SUMMARY.md
├── 003-auth-implement/
│   ├── completed/
│   │   └── 003-auth-implement.md
│   ├── code/                  # Output artifacts
│   └── SUMMARY.md
└── 004-auth-refine/
    ├── completed/
    │   └── 004-auth-refine.md
    ├── archive/
    │   └── auth-research-v1.md
    ├── auth-research.md
    └── SUMMARY.md

Naming convention:

• Folder: NNN-topic-purpose (e.g., 001-auth-research)
• Prompt: NNN-topic-purpose.md
• Output: topic-purpose.md (XML for Claude)
• Summary: Always SUMMARY.md (markdown for humans)

Creating Meta-Prompts

Determine purpose: Research, Plan, Do, or Refine?

Reference prior work: If refining or planning, reference completed research/plans:

<context>
Prior research: @ .prompts/001-auth-research/auth-research.md
</context>

Structure output for parsing: Use XML tags that next prompt can extract:

<findings>
  <finding>...</finding>
</findings>

<plan>
  <phase number="1">
    <objective>...</objective>
    <tasks>...</tasks>
  </phase>
</plan>

Create executive summary: After prompt runs, generate SUMMARY.md:

# Auth Research Summary

**Key Findings**:
- Finding 1
- Finding 2

**Next Steps**:
- Action for plan or implementation

**Full Details**: See `auth-research.md`

Chaining Prompts

Reference outputs from previous prompts:

---
name: 002-auth-plan
description: Create authentication implementation plan
---

<objective>
Based on auth research, create a detailed implementation plan.
</objective>

<research_findings>
@ .prompts/001-auth-research/auth-research.md
</research_findings>

<process>
1. Review research findings
2. Identify implementation phases
3. Define dependencies
4. Create success criteria
5. Document alternatives
</process>

The next Do prompt can then reference the plan.

Output Format Best Practices

1. XML structure - Enable efficient parsing by next Claude
2. Clear sections - Use tags like <findings>, <phases>, <decisions>
3. Metadata - Include confidence levels, timestamps, version numbers
4. Traceable - Reference source files and prior work
5. Summaries - Always create human-readable SUMMARY.md

Example: Research → Plan → Do

001-auth-research.md:

---
purpose: Research
topic: User authentication
---

<objective>
Research authentication approaches for our application.
</objective>

[Research content with XML structure]

After execution: Produces 001-auth-research/auth-research.md + SUMMARY.md

002-auth-plan.md:

---
purpose: Plan
topic: User authentication
depends: [001-auth-research]
---

<objective>
Create implementation plan based on research findings.
</objective>

<research>
@ .prompts/001-auth-research/auth-research.md
</research>

[Planning content with phase breakdown]

After execution: Produces 002-auth-plan/auth-plan.md + SUMMARY.md

003-auth-implement.md:

---
purpose: Do
topic: User authentication
depends: [002-auth-plan]
---

<objective>
Implement authentication based on plan.
</objective>

<plan>
@ .prompts/002-auth-plan/auth-plan.md
</plan>

[Implementation steps and code]

After execution: Produces code artifacts in 003-auth-implement/code/ + SUMMARY.md

Running Meta-Prompts

1. Create folder: .prompts/NNN-topic-purpose/
2. Create prompt: NNN-topic-purpose.md with frontmatter and structure
3. Store prompt: Place in completed/ after running
4. Generate output: Run prompt and save XML output
5. Create summary: Generate SUMMARY.md for human review
6. Chain: Reference output in next prompt

Reference Files

For detailed guidance:

• references/purposes.md - Deep-dive on each purpose
• references/output-formats.md - XML structure patterns
• references/examples.md - Complete examples by type
• references/chaining.md - Multi-prompt workflows